Cognition xxx (2011) xxx–xxx
Contents lists available at ScienceDirect
Cognition
journal homepage: www.elsevier.com/locate/COGNIT
Listeners invest in an assumed other’s perspective despite cognitive cost
Nicholas D. Duran ⇑, Rick Dale, Roger J. Kreuz
Department of Psychology, University of Memphis, Memphis, TN 38152, United States
a r t i c l e
i n f o
Article history:
Received 17 February 2011
Revised 3 June 2011
Accepted 13 June 2011
Available online xxxx
Keywords:
Perspective-taking
Audience design
Visuospatial processing
Mental rotation
Mousetracking
a b s t r a c t
We explored perspective-taking behavior in a visuospatial mental rotation task that
requires listeners to adopt an egocentric or ‘‘other-centric’’ frame of reference. In the current task, objects could be interpreted relative to the point-of-view of the listener (egocentric) or of a simulated partner (other-centric). Across three studies, we evaluated
participants’ willingness to consider and act on partner-specific information, showing that
a partner’s perceived ability to contribute to collaborative mutual understanding modulated participants’ perspective-taking behavior, either by increasing other-centric (Study
2) or egocentric (Study 3) responding. Moreover, we show that a large proportion of participants resolved referential ambiguity in terms of their partner’s perspective, even when
it was more cognitively difficult to do so (as tracked by online movement measures), and
when the presence of a social partner had to be assumed (Studies 1 and 2). In addition, participants continued to consider their partner’s perspective during trials where visual perspectives were shared. Our results show that participants will thoroughly invest in
either an other-centric or egocentric mode of responding, and that perspective-taking
strategies are not always dictated by minimizing processing demands, but by more potent
(albeit subtle) factors in the social context.
Ó 2011 Elsevier B.V. All rights reserved.
1. Introduction
When two people discuss the world around them, they
adapt their utterances to particular perspectives in order to
be understood. Suppose you, the reader, wish to know
where Al Green’s world-famous Full Gospel Tabernacle
church is. If the authors knew you were from Memphis,
we could respond: ‘‘By Graceland.’’ If you had a novice’s
perspective on Memphis, we might adorn the description
with more detail, such as: ‘‘About 20 min south of downtown Memphis, off Highway 55, also near Graceland.’’
(e.g., Fussell & Krauss, 1992; Isaacs & Clark, 1987). If our
spatial perspectives differed in some way (e.g., we were
chatting over the phone), I might give you directions by
tailoring them to your spatial perspective, such as ‘‘On your
left will be Graceland, then Al Green will be just. . .’’ (e.g.,
⇑ Corresponding author. Tel.: +1 909 569 2488; fax: +1 901 678 1336.
E-mail address: [email protected] (N.D. Duran).
Mainwaring, Tversky, Ohgishi, & Schiano, 2003; Schober,
1993, 1995). These adaptive processes are sensitive to a
variety of factors, and sometimes speakers mix strategies
during an interaction (Tversky, Lee, & Mainwaring, 1999).
Research conducted during the past couple of decades
has revealed a diversity of constraints on how these speech
acts are shaped by perspective (Schober, 1998, 2009). In
general, when conditions are right, and cognitive constraints are satisfied, speakers seem to employ perspectives adaptively (see also Brown-Schmidt, 2009).
Relatively less is known, however, about how the addressee herself adapts to utterances inherently involving
a spatial perspective. As noted above, speakers are neither
always careful nor explicit with perspective choice, and listeners themselves can face moments of ambiguity. Above,
when ‘‘On your left. . .’’ was used, it was assumed that the
traveler is headed south from downtown Memphis. Yet
conversational conditions do not always allow such ready
assumptions. In the construction of directions relative to
0010-0277/$ - see front matter Ó 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.cognition.2011.06.009
Please cite this article in press as: Duran, N. D., et al. Listeners invest in an assumed other’s perspective despite cognitive cost. Cognition
(2011), doi:10.1016/j.cognition.2011.06.009
2
N.D. Duran et al. / Cognition xxx (2011) xxx–xxx
some map, ‘‘on the left’’ is cognitively challenging and demands a referent object, as does ‘‘near’’ and ‘‘far’’ though
they may be less challenging (Franklin & Tversky, 1990).
Earlier studies have suggested that perspective consistency facilitates comprehension (Black, Turner, & Bower,
1979; Tversky et al., 1999), and that different perspectives
comprehended from text may allow varying levels of cognitive flexibility (Brunyé, Rapp, & Taylor, 2008). Many
studies have explored the role of particular perspectives
for memory encoding and access of maps or routes
(McNamara, Sluzenski, & Rump, 2008; Mou, Liu, & McNamara, 2009), and such memories and related inferences
are sensitive to factors such as familiarity with a terrain
or environment, and aspects of a terrain like relative object
salience (Taylor & Tversky, 1992, 1996). In addition, extensive work by Carlson and colleagues has shown that comprehenders might maintain activation across multiple
perspective types before one is selected (Carlson, 1999;
Carlson & Covey, 2005), a notion that Taylor and colleagues
confirmed using online measures and event-related potentials (Taylor, Naylor, Faust, & Holcomb, 1999).
Though this extensive work on perspective comprehension and memory has done much to reveal underlying cognitive processes, we revisit a basic experimental design to
ask the question: What spatial perspective do listeners
adopt in the face of fundamental ambiguity? The work just
reviewed almost always resolves the ambiguity for the listener in the task. For example, Taylor et al. (1999) ‘‘force’’
the participants to one or another perspective and observe
the effects on processing (see also Black et al., 1979; Lee &
Tversky, 2005; Samson, Apperly, Braithwaite, Andrews, &
Bodley Scott, 2010; Tversky et al., 1999). The research on
speakers’ choices (e.g., Schober, 1993) is more open-ended
– the tendency for speakers to adopt one or another framework in an open-ended way has been the focus of conversational spatial perspective-taking work.
To study the underlying cognitive processes involved in
open-ended listener perspective-taking, we constructed a
simple scenario modeled after Schober (1993) and Tversky
and Hard (2009), but modified it to tap into the online
comprehension of referential terms. This classic experimental design has been used frequently to study production (as reviewed above), but rarely have the
spontaneous tendencies of listeners been studied (e.g., Fitneva & Song, 2009). In such cases, do listeners assume a
consistent reference-frame over the course of an interaction? Previous research suggests that perspective switch
costs are present during comprehension (Black et al.,
1979), but in this research the listener was not given the
opportunity to spontaneously choose a perspective under
ambiguity.
In addition, will taking another person’s spatial perspective (e.g., their left) always induce cognitive difficulty?
As reviewed above, speakers sometimes choose the addressee’s perspective, and here we test the same question with
listeners. If taking another’s perspective is cognitively challenging, will participants choose it as much as they would
their own reference frame? With the cognitive expense of
taking a perspective outside of one’s own, our design permits open-ended exploration of the selection tendencies
of comprehenders to take on their own or another’s per-
spective. One could predict that cognitive load alone, without sufficient interactive constraints, favors taking the
simplest perspective (often identified as one’s own, from
Levelt, 1989 to Epley, Keysar, Van Boven, & Gilovich,
2004, though there exists no consensus).
1.1. The social perspective-taking task
We modeled our task on the design of Schober (1993),
who used a basic object-identification task to study perspective taking in speakers. Speakers saw two identical objects placed on a two-dimensional surface, and had to
induce listeners to choose one of these objects. The speaker
and listener could occupy different positions, under relative rotation of 90° or 180°, on this two-dimensional surface (designated on a simulated, drawn table on a piece
of paper). Speakers thus had to formulate utterances using
spatial descriptions that involved particular perspectives.
For example, choosing her own perspective, a speaker
could simply say: ‘‘The object to my right.’’ This perspective is often termed the egocentric perspective, because
its frame of reference is the speaker herself. The speaker
may also say ‘‘The object to your left,’’ thus taking the other
person’s perspective, which we will refer to as ‘‘othercentric.’’ Other possible descriptions can be used, such as
‘‘object-centered’’ (‘‘Choose the one nearest the airplane
picture’’). In addition, speakers may simply produce
ambiguous statements, like ‘‘The one on the left’’ when
their positions are misaligned, and leave it to the listener
to figure things out.
General findings using this design are that speakers
take the addressee’s perspective (i.e., they are othercentric) when communicating with a simulated partner,
but will coordinate perspectives with actual partners when
they are present (Schober, 1993), and will choose objectcentered descriptions at times to avoid choosing an egocentric or other-centric perspective (Schober, 1995). In
addition, they will adapt these utterances under different
task conditions, such as directing their partners to the location versus asking about the location (Mainwaring et al.,
2003). Others have found that taking the other-centric perspective may involve cognitive cost (Horton & Keysar,
1996). This simple task has thus revealed the adaptive tendencies of speakers in choosing spatial perspectives when
formulating directions.
More recently, Tversky and Hard (2009) employed this
sort of task in a simple experiment that showed that the
mere implied presence of another person shaped the spatial perspective used by survey respondents. In a singletrial study, they presented participants with a picture of
two objects and queried them about where one of the objects was located. The presence of a person in the picture,
implying an ‘‘other’’ reference frame, induced usage of
other-centric responses (e.g., ‘‘on his left’’). In a second
study, they emphasized action in the instructions, and induced further other-centric descriptions. The authors concluded that the very presence of another person results in
the use of ‘‘external’’ reference frames, and because this
was done with only minimal cuing, there is a natural and
spontaneous tendency to embody the spatial perspectives
of another’s actions or potential actions.
Please cite this article in press as: Duran, N. D., et al. Listeners invest in an assumed other’s perspective despite cognitive cost. Cognition
(2011), doi:10.1016/j.cognition.2011.06.009
N.D. Duran et al. / Cognition xxx (2011) xxx–xxx
One implication from Tversky and Hard (2009), and
from the results of Schober (1993), is that participants
are willing, even with very minimal and non-interactive
cues, to take the perspective of another, despite the fact
that some have suggested that this perspective is not the
default and is cognitively costly (Epley et al., 2004; Horton
& Keysar, 1996). However, in these previous studies, no online measure of cognitive processing was extracted to confirm either of these findings in the context of the
perspective-taking task. To do so here, we created a scenario that required participants to select and present objects to a simulated partner. The generated actions were
analyzed for overall time of execution, as well as for more
nuanced and graded signatures of difficulty that would be
problematic, if not impossible, to capture with simple button presses.
Another unaddressed issue is that neither Schober
(1993), nor Tversky and Hard (2009), explored the tendencies of the listener, who may be equally willing to invest in the other-centric perspective during occasionally
ambiguous exchanges. In the following three studies,
our task offers answers to these basic questions. First,
we show that even with a cognitively costly other-centric perspective, requiring measurable mental rotation
to take on this perspective, comprehenders of directions
will invest other-centrically. We show that this happens
even during trials when there is a shared perspective –
when the egocentric perspective is all that is needed to
solve the task – participants still pay the cognitive cost
to ‘‘transform’’ the visual array to accommodate an assumed partner. In a second and third study, we show
that comprehenders’ perspective-taking behavior is influenced by the pragmatic constraints in the communicative environment. Participants may seek to optimize
mutual understanding with their partner, and depending
on an assessment about the partner’s ability to do the
same, other-centric interpretations can increase or decrease dramatically. Like Tversky and Hard (2009), we
employed a very simple task, but embedded it in a sequence of trials and captured online cognitive measures.
And like these authors, we show that there is spontaneous, strategic investment in other-centrism despite its
difficulty.
2. Study 1: Listener’s perspective choice in an
ambiguous spatial layout
We simulate a simple interaction where a person,
standing at a table, asks another person to hand him a
folder from a set of two folders laid out on the tabletop,
as in ‘‘Give me the folder on the left.’’ Such a request would
be straightforward if both speaker and addressee were
standing side-by-side with a shared perspective of the table. However, if the speaker were to walk around the table,
now facing the addressee from the opposite side, the same
request presents a coordination problem for the addressee.
Should the interpretation be grounded in one’s own frame
of reference, or that of the other? And if addressees are unable to ask for further clarification (for example, the speaker becomes distracted or momentarily leaves the room),
3
what are the constraints in the communicative context
that might influence their decision?
The above scenario was implemented in a computerbased environment where requests for a folder were made
by a virtual speaker, whose location around a simulated table changed from trial to trial. For some of these trials the
requests were ambiguous, allowing participants to respond
either ego- or other-centrically. To make a decision on how
to respond, we hypothesize that participants are likely
considering many factors, with arguably the most important being the characteristics of the speaker and his intended meaning.
For this task, there are two speaker-related influences
that have particular relevance. First, although the virtual
speaker sounds natural from turn-to-turn, the speaker is
nevertheless delivering pre-recorded speech. As a result,
participants could react by viewing this simulated partner
as a nonentity that cannot possibly possess a unique
perspective, thus making the egocentric response the only
viable option. The recording could also fail to activate
other-centric inclinations that, with a real partner, would
normally lead to other-centric responding. Second, given
the lack of explicit feedback, there is no way of knowing
the speaker’s intent, thus there is little risk in choosing a
‘‘wrong’’ response. Therefore, it is reasonable to suspect
that the interpretation that requires the least amount of
cognitive effort, namely, the egocentric response, would
be preferred.
Yet, the limitations of the simulated partner could also
be the very basis for taking the partner’s perspective. Such
behavior would occur if participants were approaching the
interaction as a collaboration for establishing mutual
understanding. According to the work of Clark and colleagues, conversational partners optimize understanding
by attempting to minimize the effort for both themselves
and their partner (Clark, 1996; Clark & Krych, 2004; Clark
& Wilkes-Gibbs, 1986). Moreover, if it appears that one of
the partners is failing to understand (Horton & Gerrig,
2002, 2005), or is likely to find the interaction difficult
(Bortfeld & Brennan, 1997; Brennan, 1990), the other will
expend greater cognitive effort to ensure mutual understanding. These assessments of the partner are rapidly
made, and can be influenced by knowledge of the other’s
needs, characteristics, or limitations at the start of the
interaction, as in adjustments based on a partner’s age or
a known language disorder (Newman-Norlund et al.,
2009; Perkins & Milroy, 1997). Here, not only must an
assessment be made in the service of comprehension, but
it must also be extended to a partner who is incapable of
collaboration. If the participant accepts the burden of
ensuring mutual understanding, and the limitations of
the recorded speaker are deemed sufficient, we would expect higher rates of other-centric responding.
Finally, a central claim emphasized earlier is that the
other-centric response requires greater cognitive effort.
To be clear, we consider cognitive effort to be the amount
of mental work devoted to linguistic comprehension. But
unlike the many language studies where effort is confined
to retrieving words or parsing sentences, here effort is extended to a larger pragmatic and visuospatial context. To
interpret a verbal request other-centrically, not only must
Please cite this article in press as: Duran, N. D., et al. Listeners invest in an assumed other’s perspective despite cognitive cost. Cognition
(2011), doi:10.1016/j.cognition.2011.06.009
4
N.D. Duran et al. / Cognition xxx (2011) xxx–xxx
participants consider the abilities of their partner, but also
their partner’s location in space. As such, cognitive effort
can be operationalized in the same ways as those used in
well-established research on visuospatial perspective
transformations (see Zacks and Michelon (2005) for a
review).
In this previous research, it has been shown that it
takes longer to describe objects from locations that are
offset from one’s own location (i.e., egocentric). Moreover,
the greater the deviation from the egocentric perspective,
the greater the time it takes to respond. This effect also
holds when describing objects from another person’s
perspective (Schober, 1993, 1995), regardless of whether
the person is real or simulated (Amorim, 2003; Schober,
1993) or perspective choice is spontaneous or forced
(Samson et al., 2010). Our scenario allows participants
to freely adopt the perspective of a simulated partner,
and to do so at varying locations of difficulty. This difficulty is implemented by offsetting the partner’s location
by 90°, 180°, and 270° relative to the participant. The
further participants must mentally rotate from their
own egocentric perspective to ‘‘view’’ the table from their
partner’s egocentric perspective, the greater the cognitive
effort invested.
Fig. 1. The experimental interface for capturing egocentric or othercentric behavior in a single computerized task. Note: The numeric labels
are for explanatory purposes only and were not seen by participants.
time (although payment would be forfeited per Amazon
Mechanical Turk rules).
2.1. Participants
2.2. Method
We recruited 89 participants using Amazon’s Mechanical Turk.1 Mechanical Turk is an online service that allows
users to post tasks that require human expertise, like rating
the usability of websites or compiling lists of business addresses. Tens of thousands of potential ‘‘workers’’ complete
tasks on their computers and receive a nominal payment
in return. Recently, cognitive psychologists and other scientists have begun exploring the potential of Mechanical Turk
for behavioral research (e.g., Dale & Duran, in press; Dale,
Duran, & Morehead, submitted for publication; Frank &
Gibson, 2011). Respondents are found to exhibit consistently
high inter-subject reliability (Kittur, Chi, & Suh, 2008; Snow,
O’Connor, Jurafsky, & Ng, 2008; Sorokin & Forsyth, 2008),
and maintain excellent reliability and significant correlations with laboratory-collected data (Munro et al., 2010).
Moreover, because participants are not necessarily aware
that they are being tested, nor is an experimenter present,
demand characteristics are reduced.
Participants in the current study (i.e., ‘‘task’’) were paid
40 cents for contributing responses over a 7–12 min
period. To identify unique responders, the IP address of
each participant’s computer was recorded and associated
with their data. If multiple sets of data had the same IP
address, we discarded these sets because the same participants may have completed the task more than once. If IP
addresses did not originate from the United States, these
data too were discarded. Once we identified unique
responders, the IP addresses were deleted from all files.
It should also be noted that prior to committing to the
study, participants were aware of the general purpose of
the study and that they were free to terminate at any
1
www.mturk.com.
The experiment was implemented in Adobe Flash and is
accessible from an Internet web browser.2 Upon navigating
to the correct webpage, participants read brief instructions
informing them that they would have to respond to simple
requests from a simulated partner.3 The participants then
proceeded by clicking a button with their computer mouse
to reveal an image of an empty tabletop that was centered
within a 330-by-330-pixel region on their computer screen.
After a brief pause (around 500 ms), a recording of their
male partner’s voice was automatically played. The partner
directed the participant to select one of two folders that
would appear on the tabletop. Each folder was 67-by-50pixels in size. The partner’s directions always began with
the phrase ‘‘Give me the folder that is. . .’’ and was completed
with, ‘‘on the right,’’ ‘‘on the left,’’ ‘‘in the back,’’ or ‘‘in the
front.’’ After hearing these directions, participants clicked
on a ‘‘GO’’ label at the bottom of the screen (centered
65-pixels below the tabletop image). At this point, the orientation of the folders and the location of the partner and
participant became visible. The folders were arranged in
one of four possible orientations: top-right diagonally
(Fig. 1, Position 2 and 3), top-left diagonally, vertically, or
horizontally, whereas the participant’s location was always
located at a ground position of 0° (Fig. 1, Position 1). The
participant’s task was to select a folder and drag it to the
2
A version of the experiment can be accessed at: cognactive.org/
perspectiveTask.
3
This partner was labeled as either an ‘‘Employer’’ or ‘‘Employee’’ in
Studies 1 and 2. These designated labels were used to test a separate
research question on social identity. No clear and consistent patterns in
responses were found between these labels. The analyses here are based on
the combined datasets.
Please cite this article in press as: Duran, N. D., et al. Listeners invest in an assumed other’s perspective despite cognitive cost. Cognition
(2011), doi:10.1016/j.cognition.2011.06.009
N.D. Duran et al. / Cognition xxx (2011) xxx–xxx
partner’s position around the table. The partner could appear in any of the four positions of 0°, 90°, 180°, or 270°
(Fig. 1, Position 4).
The partner’s verbal directions were strategically paired
with particular folder/partner combinations to create 20
shared-perspective and 20 critical trials. It is during the
critical trials that a participant’s current spatial perspective
taking is revealed. For example, based on the layout of
Fig. 1, if the participant hears, ‘‘Give me the folder on the
left,’’ and selects the folder at Position 2, they are acting
egocentrically (‘‘folder on my left’’). But if the folder at Position 3 is selected, the participant is now acting ‘‘othercentrically’’ (‘‘folder on your left’’). Given the same layout,
a shared-perspective trial would occur if the participant
now hears, ‘‘Give me the folder in the front.’’ Regardless
of whether the participant takes an egocentric or othercentric perspective, the folder at Position 2 is aligned with
both perspectives and should always be selected. These trials serve as a type of control.
As participants moved toward a folder, we were rapidly
sampling the x, y coordinates of their mouse cursor (similar
to Brennan, 1990). These coordinate positions were recorded every 25 ms (a sampling rate of 40 Hz) from the
moment the ‘‘GO’’ region was clicked to the initial selection of a folder. The ‘‘GO’’ region is a point of calibration
that ensures all responses begin at the same location. The
subsequent temporal and trajectory analyses are based
on these movements.
In presenting the stimuli to participants, we randomly
shuffled shared-perspective and critical trials, ensuring
that no more than three trials of the same type ever occurred in a sequence. In each list, all possible combinations
of folder orientation and partner location were seen at
least twice – except for horizontal and vertical orientations
when participants were located at 90° or 270°. At these
combinations, the directions are no longer ambiguous
and are likely to bias perspective choice. For example,
imagine the scenario where the speaker is offset by 90°
and the folders are laid out vertically. If the speaker asks
for the folder on the right, the participant is forced to consider the speaker’s perspective because ‘‘right’’ is interpretable only from where the speaker is located. Additional
constraints made it impossible for the speaker in critical
trials (nonshared-perspective) to appear at the 0 degree
location (shared-perspective), and for the speaker in
shared trials to appear at the 180 degree location (nonshared-perspective). In the end, the 20 critical trials consisted
of 4 trials each at the 90 and 270 degree position, and 12 at
the 180 degree position. The same was true for the 20
shared-perspective trials, except now with 12 trials at
the 0 degree position. Finally, for trials at the 90 and 270
degree positions, and separately at the 0 and 180 degree
positions, there were an equal number of folders oriented
top-right and top-left diagonally, and from each direction
type. Trials at the 0 and 180 degree position also included
the horizontal and vertical orientations.
5
and excessive delays. Guessing is evident in the sharedperspective trials where a response to the partner’s directions should always converge on a single folder. If this target folder was not selected, then the response was flagged
as an error. Furthermore, if a trial lasted for more than ten
seconds, it too was flagged. Participants who made two or
more errors were excluded. A total of eight participants, or
8.98% overall, were removed. In addition, 101 trials, or
3.16% overall, were removed because the response time
to select a folder exceeded three standard deviations above
the mean (i.e., greater than 3986 ms).
2.3.1. Patterns of egocentric and other-centric responders
Based on the consistency of responses across critical
trials, participants were classified as egocentric, other-centric, or mixed responders. To identify category membership, we computed proportions of egocentric and othercentric trials for each participant. If proportion scores exceeded 0.70 for one of the two response categories, then
the participant was considered to be a member of that category. Participants scoring below 0.70 were labeled as
mixed responders. There were only 8 (or 9.76%) mixed
responders. Participants tended to be on the extreme ends
of a bimodal distribution, with 31 participants (38.80%)
classified as egocentric and 43 participants (52.44%) as
other-centric. However, this difference was not statistically
significant, {2(1) = 1.95, p = 0.16.
2.3.2. Greater processing costs in taking another’s perspective
In this analysis, we evaluate whether other-centric
responders are mentally rotating to their partner’s visual
perspective. To do so requires the participants to rotate from
their own spatial frame of reference, located at ground position 0°, to the partner’s perspective that is located at position
90°, 180°, or 270°. This rotation should be increasingly difficult the further the rotation is from the personal frame of
reference (Michelon & Zacks, 2006; Miller & Johnson-Laird,
1976; Presson & Montello, 1994; Shepard & Metzler, 1971;
Zacks, Vettel, & Michelon, 2003). Thus, when the partner is
rotated 180° (directly across from the participant) there will
be greater processing difficulty than when the partner is rotated 90° (at a ground position of 90° or 270°). Because this
behavior is expected to occur just with other-centric
responders, processing times at each of the ground positions
should also be much higher when compared to egocentric
responders.
We used a linear mixed-effects model4 to evaluate response time differences across degree of rotation (withinsubjects factor; three levels: 0° vs. 90° vs. 180°) and responder type (between-subjects factor; two levels: consistent
egocentric vs. consistent other-centric). It is important to
note that the rotated position at 90° also includes trials at
270° because both positions require equivalent degrees of
rotation. To simultaneously account for random variance
due to participants and items, ‘‘participant’’ was added as
a random effect, as well as folder orientation (e.g., horizontal, diagonal, and vertical), and verbal directions
2.3. Results
To ensure that participants were committed to the task,
we identified suspicious behavior like random guessing
4
Models were built using the PASW (SPSS) MIXED procedure, with
denominator degrees of freedom estimated using Satterthwaite’s approximation (default).
Please cite this article in press as: Duran, N. D., et al. Listeners invest in an assumed other’s perspective despite cognitive cost. Cognition
(2011), doi:10.1016/j.cognition.2011.06.009
6
N.D. Duran et al. / Cognition xxx (2011) xxx–xxx
Fig. 2. The critical trial response times and standard errors of egocentric and other-centric responders at 90 and 180 degrees of partner rotation, and
shared-perspective response times at 0 degrees of rotation. For each increase in rotation, other-centric responders take much longer than egocentric
responders. Other-centric responders are also slower than egocentric responders at each degree of rotation (Section 2.3.2), including shared-perspective
response times at 0 degrees of rotation and 90 degrees of rotation (Section 2.3.4). Furthermore, other-centric critical and shared-perspective response times
at 90 degrees of rotation are nearly identical (Section 2.3.4). Note: Statistical significance is coded as: ⁄p < .05, ⁄⁄p < .01, ⁄⁄⁄p < .001.
(‘‘e.g., ‘‘right,’’ ‘‘left,’’ ‘‘front,’’ and ‘‘back’’). As an additional
precaution that would control for practice effects, models
with and without trial number were compared using a
log-likelihood ratio test as recommended by Baayen (2008)
and Quene and van den Bergh (2008). The addition of trial
number resulted in a model with a better overall fit, and
was therefore included as another random effect.5
The overall model revealed a statistically significant
main effect for degree of rotation, F(2, 1424.03) = 48.62,
p < .001, a significant main effect for responder type
F(1, 66.75) = 7.72, p = .007, and a significant interaction between the two factors, F(2, 2066.31) = 2066.31, p = .001.
Given the significant interaction, follow-up simple main
effects were conducted across degree of rotation for egocentric and other-centric responders. For other-centric
responders, there was an overall significant effect,
F(2, 1562.86) = 54.02, p < .001, resulting from increased
cognitive processing across degree of rotation (see Fig. 2).
Planned comparisons show significant difference between
0° and 90° (mean difference = 172 ms6), p < .01, Cohen’s
d = .17; and between 90° and 180° (mean difference = 164.66 ms), p < .01, d = .27. For egocentric responders,
although no significant increase was expected, these participants also appear to consider some aspect of the partner’s
5
These random effect variables are included in all subsequent mixedeffects analyses. The specific model used here was chosen to give a
conservative unbiased evaluation of the dependent variables, following
established parameter specification techniques (Baayen, 2008; Baayen,
Davidson, & Bates, 2008; Quene & van den Bergh, 2008).
6
All reported and plotted means are the estimated marginal means that
have been adjusted by linear mixed-effects models to remove random
variance due to participant, item, and practice effects. The estimated
marginal means are those used by the model to derive inferential statistics,
and represent a better, unbiased account of outcome response patterns.
changing location, F(1, 1792.42) = 9.12, p < .001, but only
when location shifted from 0° to 90° (mean difference = 123 ms), p < .01, Cohen’s d = .05. Yet, it is clear that
other-centric responders are engaged in greater cognitive
work, as the average linear increase of response time for
other-centric responders was much higher than that of egocentric responders. To demonstrate this overall increase, an
additional mixed-effects model was computed with response time regressed on degree of rotation, now coded as
a continuous factor (e.g., ‘‘1’’, ‘‘2’’, and ‘‘3’’). The slope of
the linear regression line for other-centric responders,
b = 170.93, is much steeper than that of egocentric responders, b = 75.67, F(2, 1637.83) = 65.69, p < .001.
The other-centric response times at each of the three
ground positions (0°, 90°, and 180°) were also significantly
higher than those of egocentric responders. A series of
analyses based on the overall model confirmed that at 0
degrees of rotation, other-centric responders were
183 ms slower, F(1, 78.36) = 3.57, p = .06 (marginal),
d = 0.37; at 90 degrees of rotation, other-centric responders were 232 ms slower, F(1, 92.07) = 5.29, p = .02,
d = 0.40; and at 180 degrees of rotation, other-centric
responders were 361 ms slower, F(1, 79.06) = 13.81,
p < .001, d = 0.50. Taken together with the previous results,
even though egocentric responders appear to be affected at
some level by the locational changes of the recorded
speaker, other-centric responders are taking on the additional processing costs of mental rotation.
2.3.3. Greater processing complexity and competition for
other-centric responders
As participants use their mouse cursor to select a
folder, fine-grained changes in the decision movement
Please cite this article in press as: Duran, N. D., et al. Listeners invest in an assumed other’s perspective despite cognitive cost. Cognition
(2011), doi:10.1016/j.cognition.2011.06.009
7
N.D. Duran et al. / Cognition xxx (2011) xxx–xxx
Table 1
Study 1: Means and standard errors for directional shifts, maximum deviation, and acceleration components at 0, 90, and 180 degrees of rotation.
Rotation
0
90
180
Directional shifts
Maximum deviation
Acceleration components
Ego
Other
Ego
Other
Ego
Other
1.23 (.06)
1.32 (.07)
1.36 (.07)
1.47 (.05)
1.65 (.07)
2.07 (.07)
12.16 (1.58)
11.12 (1.46)
8.45 (.77)
18.83 (2.27)
25.64 (2.16)
23.15 (1.54)
8.78 (.23)
9.79 (.34)
9.70 (.29)
10.29 (.23)
11.22 (.30)
12.17 (.29)
are captured. These changes reflect processing complexity
and competition as a decision is made. Here, we consider
three trajectory measures of complexity and competition:
directional shifts in motion, maximum deviation, and acceleration components. Directional shifts occur when the
mouse cursor switches direction along the x- or y-axis,
depending on the orientation of the folders. For example,
as a participant moves from the bottom of the screen to a
folder, they might change their mind and move towards
the other folder. This change will be reflected in a flip
of direction along the axis in which folders are arranged
(x-axis for diagonal and horizontal; y-axis for vertical).
We can also measure the strength of the initial bias that
brought about the directional flip. Because the visually
co-present folders in the horizontal and diagonal configurations are laid out side-by-side along the x-axis, and
the initial location of the cursor at the bottom of the
screen is set equidistantly between the two folders,
(x, y = 0, 0), we can measure the maximum deviation
away from 0 that the trajectory is ‘‘pulled’’ (also see Spivey & Dale, 2006). Here, ‘‘pull’’ is toward the competitor
folder, as would be the case for other-centric responders
that move towards the ‘‘egocentric’’ folder while making
a final selection. Presumably, the further responders
travel towards the egocentric folder en route to their
final selection, the stronger the bias to respond egocentrically (or other-centrically in the case of egocentric
responders).
Alternatively, uncertainty and competition in selecting
a folder can occur without directional change along an axis.
In this scenario, a participant makes an initial confident response towards one folder, but slows down, and then
speeds up, as they deliberate about their final response.
The acceleration component measures the number of such
response hesitancies (Dale & Duran, in press). All three
measures supplement reaction time by providing further
evidence of the increased cognitive effort associated with
being other-centric.
2.3.3.1. Directional shifts. As with the reaction time data, a
mixed-effects analysis was conducted with degree of rotation and responder type as fixed effects. Both main effects
were statistically significant (both at the p = .001 level), as
was the interaction term F(1, 1938.59) = 8.55, p < .001.
Based on the means reported in Table 1, the directional
shifts for other-centric responders appear to increase with
degrees of rotation. This pattern was validated with a simple effects analysis, F(2, 1418.71) = 31.97, p < .001, with
significant increases from 0° to 90° (mean difference = 0.18
shifts), p = .04, d = .15 and from 90° to 180° (mean difference = 0.42 shifts), p < .001, d = .32. For egocentric respond-
ers, there were no statistically significant increases.
Further comparisons also show that the other-centric
responders had more shifts than egocentric responders at
each of the three ground positions: 0° (mean difference = 0.24 shifts), F(1, 103.19) = 3.03, p = .08 (marginal),
d = .27; 90° (mean difference = 0.33 shifts), F(1154.83) =
4.83, p = .03, Cohen’s d = .36; and 180° (mean difference = 0.71 shifts), F(1, 106.38) = 26.43, p < .001, Cohen’s
d = .49.
2.3.3.2. Maximum deviation. The main effect for responder
type was significant, p < .001, as well as a significant interaction between responder type and degree of rotation,
F(2, 1693.31) = 3.41, p = .03. For other-centric responders,
there was greater movement toward a competitor folder
between 0 and 90 degrees of rotation (mean difference =
6.81 pixels), p = .004, d = .17. For egocentric responders,
there were no equivalent movements. The comparisons
at each of the ground positions also show that othercentric responders deviate more toward the competitor
folder
at
0°
(mean
difference = 6.67
pixels),
F(1, 153.06) = 4.31, p = .04, Cohen’s d = .17; at 90° (mean
difference = 14.52 pixels), F(1, 218.63) = 16.89, p < .001,
d = .48; and at 180° (mean difference = 14.71 pixels),
F(1, 83.19) = 10.83, p < .001, d = .62.
2.3.3.3. Acceleration components. Both main effects were
statistically significant (degree of rotation, p < .001; responder type, p = .01), as was the interaction term,
F(2, 1937.42) = 3.11, p = .04. Follow-up analysis on the
interaction shows an increase for other-centric responders
across each degree of rotation, F(2, 1740.46) = 4.89,
p = .008; with the planned comparison between 0° and
90° being significant (mean difference = 0.93 components),
p = .004, d = .10, and the comparison between 90° and 180°
also significant (mean difference = .95 components),
p = .004, Cohen’s d = .18. We also found a significant increase of acceleration components for egocentric responders, F(2, 1569.07) = 21.59, p < .001, but only for the planned
comparison between 0° and 90°, mean difference = 1.00
components, p = .01, d = .12. Finally, other-centric responders exhibited more acceleration components at each of the
three ground positions: 0° (mean difference = 1.51 components), F(1, 82.09) = 4.06, p = .05, d = .37; 90° (mean difference = 1.43 components), F(1, 100.54) = 3.32, p = .07
(marginal), d = .32; and 180° (mean difference = 2.47 components), F(1, 83.19) = 10.83, p = .001, d = .43.
Overall, the trajectory measurements provide a more
detailed evaluation of the decision dynamics that occur
during the time required to initiate and finalize a response
movement. For each measure, the decision to be other-
Please cite this article in press as: Duran, N. D., et al. Listeners invest in an assumed other’s perspective despite cognitive cost. Cognition
(2011), doi:10.1016/j.cognition.2011.06.009
8
N.D. Duran et al. / Cognition xxx (2011) xxx–xxx
centric is marked by greater instability, and that the activation of the competitor response, e.g., the egocentric option,
is strongest for other-centric responders. Moreover, the
processing difficulty associated with mentally rotating to
a partner’s visual perspective is amplified the further
respondents must travel in the rotation. This latter finding
generally holds for other-centric response behavior, with
the exception of maximum deviation from ground positions 90° to 180° (although the increase from 0° to 180°
was statistically significant, p = .01). And like the overall
response time analysis, there was some evidence that the
partner’s changing location added some complexity to egocentric responding, but only with acceleration components
between 0 and 90 degrees of rotation. There were no
equivalent effects with directional shifts and maximum
deviation.
2.3.4. Perspective-taking during shared trials
In the previous response time analysis (Section 2.3.2
above), the 0-degree rotated position was included as a
baseline in which participant and their partner shared
the same view of the table (and consequently, visual perspectives were aligned). At 0° there is no need to mentally
rotate; thus, other-centric response times may be similar
to that of egocentric responders. However, as evident by
the 183 ms difference, other-centric responders are somewhat slower. This result suggests other-centric responders
are continuing to consider their partner’s perspective, even
when it is unnecessary to do so. As further evidence, we
compared the shared-perspective trials of egocentric and
other-centric responses at 90 degrees of rotation. If
other-centric responders are mentally rotating and egocentric responders are not, then other-centric responders
will again be slower. Indeed, this result was confirmed,
(mean difference = 207 ms); F(1, 90.44) = 5.24, p = .02,
d = .29 (see Fig. 2). Finally, we assessed just other-centric
responses during shared-perspective and critical trials
at 90 degrees of rotation. Unlike the critical trials,
the shared-perspective trials present a situation where
the perspectives of participant and partner converge on
the same folder. These trials may be easier because there
is no conflict between perspectives (‘‘your right’’ or ‘‘my
right’’), thus there is no need for other-centric participants
to inhibit activation from their own perspective (‘‘my
right’’). However, the response times for critical and
shared-perspective trials were nearly identical, (shared:
M = 1756 ms,
SE = 44 ms;
critical:
M = 1792 ms,
SE = 38 ms); p > .10, suggesting that other-centric responders are continuing to mentally rotate, even when the visual
context does not place any demands on them to do so.
2.4. Brief discussion
Given an ambiguous spatial perspective task, participants exhibit an almost equivalent preference in taking
an egocentric or other-centric perspective. This commitment to any one perspective is a consistent mode of
responding that persists over an extended interaction with
an assumed social partner. Furthermore, we show that
other-centric responders are spontaneously choosing what
is a more cognitively challenging behavior. To take another’s perspective, responders are mentally rotating to
the spatial location of their partner. In doing so, they are
quite literally taking on the visual perspective of their partner. Not only does this process take longer the greater the
required rotation, but it is also marked by signatures of
processing complexity. Even so, more than half of the participants in this study chose to accommodate the perspective of their partner – a partner that participants know to
be a mere simulation.
3. Study 2: Perspective-taking modulated by
attributional cues
To respond other-centrically, participants appear to be
engaging in a pattern of behavior that optimizes the understanding of the other. Although the ‘‘other’’ in the current
context has no need or ability to understand, participants
still adhere to principles of collaboration, an apparent default for social interaction. Thus, when a partner is unable
to contribute equally to the collaboration, the more adept
partner expends greater effort to do what the other cannot
(Branigan, Pickering, Pearson, & McLean, 2010; Brennan,
1990). Of course, the question still remains: why then do
a large proportion of participants still act egocentrically?
One possibility is that the constraints in Study 1 are simply
insufficient. These constraints lack the force to trigger
attributions that are necessary for other-centric responding (Clark & Schaefer, 1989; Clark & Wilkes-Gibbs, 1986;
Wilkes-Gibbs & Clark, 1992). Such ineffectiveness might
be due to responders who have a high criterion for establishing a collaboration or noticing when a collaboration becomes unbalanced. Either way, by introducing a potentially
more relevant constraint, or attributional cue, the rate of
other-centric responding should dramatically increase.
However, if responders are ignoring the other as irrelevant,
and acting solely to minimize their own cognitive effort,
then the additional cue should have little effect on the
overall pattern of perspective-taking behavior.
In Study 2, we chose an additional attributional cue that
would be relevant only if participants were treating the
simulated speaker as a potential collaborator. As such, we
simply informed participants that their partner does not
know where they are seated at the table. Now, when the
participant hears a verbal direction like, ‘‘Give me the
folder on the right,’’ the interpretation must take into account that the partner’s knowledge state precludes the
possibility of ‘‘on the right’’ meaning the participant’s
‘‘right’’. The partner can only give directions from her
own perspective; thus, the participant should respond
other-centrically. In addition, using the same cognitive
measures, we consider whether this additional cue to be
other-centric makes an other-centric response easier, or
continues to be the more difficult option.
3.1. Participants
We recruited 96 participants using Amazon’s Mechanical Turk. Each was compensated with a small monetary
Please cite this article in press as: Duran, N. D., et al. Listeners invest in an assumed other’s perspective despite cognitive cost. Cognition
(2011), doi:10.1016/j.cognition.2011.06.009
N.D. Duran et al. / Cognition xxx (2011) xxx–xxx
payment (40 cents). The participants’ IP addresses were
also different from those in Study 1.
9
responders increased by 22%, {2(1) = 6.82, p < .01, whereas
egocentric responders decreased by 16%, {2(1) = 4.21,
p = .04.
3.2. Method
The only difference from Study 1 is a slight change in
the initial instructions. We now include the statement:
‘‘Your partner DOES NOT know where you are seated.’’
All other aspects of the experiment were identical.
3.3. Results
We removed 15 participants, or 15.62% overall, who appeared to be randomly guessing on shared-perspective trials, or took an excessive amount of time to answer (greater
than 10 s). We also removed 61 trials, or 2.62% overall, that
were three standard deviations above the response time
mean for selecting an initial folder (i.e., 3634 ms).
3.3.1. Patterns of egocentric and other-centric responders
If participants are considering what their partner
knows, it is expected that the introduction of a more perspicuous attributional cue will shift the distribution of consistent responders toward other-centric. Indeed, there was
evidence of this effect. Based on the proportion of response
type over critical trials, 18 (22%) participants consistently
responded egocentrically on more than 70% of the trials,
and 60 (74%) participants consistently responded othercentrically. This difference between other-centric and egocentric responders was significant, {2(1) = 22.62, p < .001.
Only 3 (3.7%) participants were mixed responders (see
Fig. 3). Furthermore, not only was there a difference of
other-centric and egocentric responding within Study 2,
but when compared against Study 1, other-centric
3.3.2. Greater processing costs in taking another’s perspective
A linear mixed-effects model wth response time as the
dependent variable revealed a significant interaction between rotation and responder type, F(2, 2193.62) = 7.19,
p = .001. Starting with a closer examination of othercentric responders, the linear increase across 0°, 90°, and
180° was statistically significant, F(2, 1061.13) = 44.19,
p < .001. Planned comparisons show a marginally significant increase between 0° and 90° (mean difference = 59 ms), p = .07, d = .10; and between 90° and 180°
(mean difference = 202 ms), p < .001, d = .32 (see Fig. 4).
For egocentric responders, there was an overall marginally
significant increase, F(2, 2019.48) = 2.42, p = .09, but was
confined to the increase from 0° to 90°, mean difference = 120 ms, p = .04, d = .13. To evaluate whether this
egocentric increase was equivalent to the other-centric increase, we regressed reaction time on degree of rotation,
with degree of rotation coded as a continuous variable.
As expected, the fitted regression lines were significantly
different, with other-centric rising on average 132 ms
and egocentric only rising 47 ms, F(2, 1904.15) = 44.70,
p < .001.
Interestingly, in the between-subjects analysis at each
degree of rotation, other-centric responders took much
longer to respond than egocentric responders at two of
the three ground positions (Fig. 4). Follow-up comparisons
show statistically significant differences at 0° (mean difference = 203 ms), F(1, 88.01) = 2.91, p = .09 (marginal),
d = .28; and at 180° of rotation (mean difference = 380 ms),
F(1, 90.35) = 9.99, p = .002, d = .46.
3.3.3. Processing complexity and competition for other-centric
responders
Separate mixed-effects models were built with each of
the three trajectory measures as dependent variables. The
measures, and their descriptives, are displayed in Table 2.
3.3.3.1. Directional shifts. The interation for the overall
model was statistically significant, F(2, 2094.61) = 3.60,
p = .03. With this measure, other-centric responders had
more shifts across rotated positions, F(2, 2092.19) = 19.85,
p < .001, particularly between 0 and 90 degrees rotation,
mean difference = .341 shifts, p < .001, d = 28. And, as in
the previous analysis, other-centric responders had more
shifts than egocentric responders at each ground position:
at 0° (mean difference = .494 shifts), F(1, 108.16) = 8.81,
p = .004, d = .42; at 90° (mean difference = .449 shifts),
F(1, 150.56) = 6.13, p = .01, d = .36; and at 180° (mean difference = .796 shifts), F(1, 115.24) = 22.15, p < .001, d = .58.
Fig. 3. Distribution of egocentric and other-centric responders across the
three studies. For Study 2, there are a greater number of other-centric
responders and fewer egocentric responders when compared to Study 1.
For Study 3, this pattern is reversed, with a greater number of egocentric
responders and fewer other-centric responders when compared to Study
2 and Study 1.
3.3.3.2. Maximum deviation. Only a statistically significant
main effect for degree of rotation was found for this study,
F(2, 1827.93) = 4.19, p = .02, d = .25, showing a significant
increase of 6.73 deviations from 0 to 90 degrees of rotation.
3.3.3.3. Acceleration components. A statistically significant
main effect for degree of rotation, F(2, 2092.22) = 2.27,
Please cite this article in press as: Duran, N. D., et al. Listeners invest in an assumed other’s perspective despite cognitive cost. Cognition
(2011), doi:10.1016/j.cognition.2011.06.009
10
N.D. Duran et al. / Cognition xxx (2011) xxx–xxx
Fig. 4. The critical response times and standard errors of egocentric and other-centric responders at 90 and 180 degrees of partner rotation, and sharedperspective response times at 0 degrees of rotation. Between 0–90°, and 90–180°, other-centric responders take more time, whereas egocentric responders
increase from 0° to 90° (Section 3.3.2). However, other-centric responders are overall slower than egocentric responders (Section 3.3.2), including sharedperspective response times at 0 degrees of rotation (Section 3.3.4). Furthermore, other-centric critical and shared-perspective response times at 90 degrees
of rotation are nearly identical (Section 3.3.4).
Table 2
Study 2: Means and standard errors for directional shifts, maximum deviation, and acceleration components at 0, 90, and 180 degrees of rotation.
Rotation
0
90
180
Directional shifts
Maximum deviation
Acceleration components
Ego
Other
Ego
Other
Ego
Other
1.13 (.07)
1.22 (.09)
1.21 (.09)
1.63 (.05)
1.67 (.57)
2.01 (.56)
14.28 (3.66)
17.15 (4.65)
17.22 (4.39)
17.84 (1.46)
28.44 (2.47)
25.46 (1.59)
9.15 (.34)
9.77 (.43)
9.96 (.35)
11.24 (.23)
11.25 (.26)
12.72 (.25)
p < .001, revealed an increase from 90° to 180° (.32 components), d = .27. Also, a statistically significant main effect
for responder type, F(1, 72.44) = 6.72, p = .01, verified that
other-centric responders overall had 2.11 more components than egocentric responders, d = .48.
3.3.4. Perspective-taking in shared trials
The 0-degree rotated position represents shared perspectives where there is no need to mentally rotate. Despite this, other-centric responders still engage in greater
cognitive work, as they are 203 ms slower than egocentric
responders (see Section 3.3.2 above). When comparing the
shared-perspective trials of egocentric and other-centric
responses at 90 degrees of rotation, again, other-centric
responders are slower by 172 ms; however, this effect is
not statistically significant (Fig. 4). But importantly, the
other-centric response times during shared-perspective
and critical trials at 90 degrees of rotation did not vary.
Similar to Study 1, this result suggests that other-centric
responders are consistent across trial types, apparently engaged in a stable strategy of considering the perspective of
their partner.
3.4. Brief discussion
When the attributional cues in a communicative context are made more obvious, participants use this information to shape their perspective-taking choice. Here, a
participant is told that a partner does not know their location (and thus must infer that the partner is unable to take
their perspective). This information was sufficient to bias
nearly two-thirds of the participants to an other-centric
mode of responding. By doing so, participants are taking
on the visuospatial perspective of their partners in assessing how to optimally respond. Furthermore, this mode is
sustained throughout the trials, despite the finding that
this behavior is comparatively more difficult than egocentric responding, as it was in Study 1 (based on response
time and complexity measures).
4. Study 3: Perspective-taking under the assumption
that the direction-giver is ‘‘real.’’
Thus far, we have argued that participants adhere to
principles of collaboration when interacting with a simulated partner. Moreover, fundamental attributions about
Please cite this article in press as: Duran, N. D., et al. Listeners invest in an assumed other’s perspective despite cognitive cost. Cognition
(2011), doi:10.1016/j.cognition.2011.06.009
N.D. Duran et al. / Cognition xxx (2011) xxx–xxx
the speaker’s inability to collaborate tend to lead to greater
other-centric responding. In Study 3, we examine this
claim in greater detail by now asking: what if participants
were to interact with a seemingly real human partner? To
achieve this pretense, we use the same experimental stimuli as in Study 1, but add a short, interactive introduction
to convince participants that they are connected to a real
human being.
In this new scenario, participants have a potentially collaborative partner who can be held responsible for ensuring mutual understanding. And because the partner is
initiating the request, and is the one that wants the folder,
the burden of collaboration is likely shifted to the speaker
– at least to a greater extent than in Study 1 and 2. In these
previous studies, there was no way for the recorded speaker to know the participant’s perspective, by virtue that the
speaker was removed in space and time. And because they,
the participants, were the only ones able to adopt an othercentric perspective, they seemingly obliged. In this study
we remove the partner’s previous limitations, and expect
the new attributional information to lead to greater egocentric responding.
4.1. Participants
Eighty participants were recruited from Amazon’s
Mechanical Turk and were paid 75 cents for their participation. The increase in payment is due to an increase of
about 5 min needed to complete the introductory stage
of the study. The participants’ IP addresses were also different than those in Studies 1 and 2.
4.2. Method
The major addition from Study 1 was to lead participants into thinking that they were remotely connecting
to another Mechanical Turk ‘‘worker’’ or Mechanical Turk
‘‘requester’’ (counterbalanced across participants). To
establish this ruse, we began with slight modifications in
how the study was advertised on Mechanical Turk, as well
as in the wording of the initial instructions. Both were altered to provide a cover story that we (i.e., the ‘‘Requesters’’) were testing software that would allow users to
engage in real-time audio communication within a Flashbased environment. However, because we were in ‘‘the
early stages of development,’’ the audio could only be
transmitted one-way. Nevertheless, we were most interested in the ability to connect two users, and to have these
users perform a joint activity within a shared virtual space.
And identical to Study 1, the instructions went onto explain that the activity would require participants to move
simulated folders in response to a direction-giver’s verbal
requests. But unlike Study 1, before participants could proceed, they had to first engage in an extended process of
connecting to what would appear to be a ‘‘real-life’’
partner.
In this second step of the introductory stage, participants were directed to press a large button labeled: ‘‘CLICK
TO CONNECT.’’ Upon clicking, participants were taken to a
screen with the message ‘‘Connecting to another Turk user,
please wait a few seconds. . .’’ or ‘‘Connecting to a Reques-
11
ter, please wait a few seconds. . .’’ displayed near the top.
The ellipses in the message above flashed every 500 ms
to mimic dynamic updating as the software attempted to
‘‘connect’’ to an outside worker or requester. After approximately 5000 ms, the ellipses stopped blinking and a new
screen appeared with the word ‘‘CONNECTED’’ in the upper
right-hand corner, and with an empty text box in the middle of the screen. After a brief delay, the male voice in the
previous experimental phases of Studies 1 and 2 began to
speak in the same tone and volume as in the experimental
trials. This male partner reiterated information from the
written instructions, which was that he alone could transmit audio but was unable to receive audio from the participant. Next, to (ostensibly) ensure that there was indeed a
connection, the male partner then asked the participant to
type the code word ‘‘CAT’’ into the empty text box and
press the ENTER key. If an incorrect code word was given,
the partner repeated the code word with a slightly different intonation. Once the male partner ‘‘received’’ the correct code word, thereby indicating to the participant that
the partner could hear him or her, the partner stated that
the instructions allowed both of them to proceed. The participant then clicked a button to begin the identical experimental task employed in Studies 1 and 2.
At the completion of the experimental task, several follow-up questions were asked to ensure that the participants believed that they had been connected to a real-life
Turk worker or requester. The first and second questions
were open-ended responses, with the first question
addressing general comments and feedback about the
Mechanical Turk task, and the second question asking,
‘‘How did you feel about your interaction with your partner?.’’ The third and fourth questions were yes or no responses, asking, ‘‘Did you FEEL like you were connected
to an actual person?,’’ and ‘‘Did you BELIEVE you were connected to an actual person?.’’ Each question was presented
in isolation on separate screens.
4.3. Results
We discarded trials from participants who responded
that they did not feel/believe that they were connected
to a real person, or if a participant indicated in any way
that they thought their partner was simulated (e.g., ‘‘it
sounded like a recording,’’ ‘‘the person was a fake’’). We
achieved a 75% deception rate in this modified task, with
60 out of the 80 participants seemingly convinced that
the virtual confederate was indeed a real person.
There were no participants who appeared to be randomly guessing, but one participant did have several trials
that exceeded the maximum time limit of ten seconds. This
participant was removed. Furthermore, 60 trials, or 2.58%
overall, were removed because they exceeded three standard deviations above the response time mean for selecting an initial folder (i.e., 3892 ms).
4.3.1. Patterns of egocentric and other-centric responders
The introduction of a seemingly real partner should
change the collaborative attributions that can be made.
In the earlier studies, the partner was an ineffectual collaborator, forcing participants to take on the additional work
Please cite this article in press as: Duran, N. D., et al. Listeners invest in an assumed other’s perspective despite cognitive cost. Cognition
(2011), doi:10.1016/j.cognition.2011.06.009
12
N.D. Duran et al. / Cognition xxx (2011) xxx–xxx
of being other-centric. Now, the partner is capable of carrying the load, allowing participants to delegate work to the
partner. In this study, there were 38 (64%) egocentric
responders, 17 (29%) other-centric responders, and 4 (7%)
mixed responders (based on 70% or more of trials being
consistently ego- or other-centric). Thus, there is a higher
rate of egocentric responding, {2(1) = 8.02, p = .004, suggesting participants were indeed sensitive to the new attributional cues. In comparison, the 20 participants who
reported having a simulated partner, and therefore less
influenced by the attributional information, were also less
likely to respond egocentrically, with a response rate that
decreased by 22%.
Lower rates of egocentric responding were also evident
in Studies 1 and 2. Compared to the current result, there
was a 27% decrease of egocentric responding in Study 1,
{2(1) = 6.93, p = .008; and a 42% decrease for Study 2,
{2(1) = 20.55, p < .001. Conversely, in comparing rates of
other-centric responding across studies, there was a 24%
increase for Study 1, {2(1) = 6.87, p = .009; and a 45% increase for Study 2, {2(1) = 20.62, p < .001.
4.3.2. Greater processing costs in taking another’s perspective
A significant main effect was found for degree of rotation and responder type, both at the p < .001 level. Importantly, the interaction term was also significant,
F(2, 1594.32), p = .004 (see Fig. 5). Simple main effects for
degree of rotation at egocentric and other-centric responders showed statistically significant increases of response
time for both responder types across degree of rotation:
other-centric responders, F(2, 1594. 61) = 28.16, p < .001,
and egocentric responders, F(2, 1593.66) = 14.74, p < .001.
These increases occured from 0° to 90°, with other-centric
responders 208 ms slower, p < .001, d = .25; and egocentric
responders 78 ms slower, p = .03, d = .08. There were also
increases from 90° to 180°, with other-centric responders
167 ms slower, p = .004, d = .25; and egocentric responders
99 ms slower, p = .007, d = .18. Although both egocentric
and other-centric responders showed an upward trend
across degree of rotation, the cognitive costs of mental
rotation enacted a greater cost on other-centric responders. This is evident in the comparison of the fitted regression lines, whereby other-centric responders had a
steeper rise in response times, b = 195 ms, than that of
the egocentric response times, b = 88 ms, F(2, 1353.19) =
48.73, p < .001.
And lastly, simple main effects for responder type at
each ground position also show that other-centric
responders required more time to respond at 0° (mean difference = 339 ms), F(1, 63.15) = 9.51, p = .003, d = .64; at
90° (mean difference = 468 ms), p < .001, d = .76; and at
180° (mean difference = 537 ms), p < .001, d = .77.
4.3.3. Greater processing complexity and competition for
other-centric responders
4.3.3.1. Directional shifts. The main effect for degree of rotation was statistically significant, F(2, 1541.61) = 6.96,
p = .001, as was the main effect for responder type,
F(1, 51.51) = 4.82, p = .03. Thus, the increase from 0° to
90° (.08 shifts) and from 90 to 180 (.161 shifts) showed
increasing difficulty that is similar for egocentric and
other-centric responders; and, overall, the other-centric
responders still had more shifts then egocentric responders
(see Table 3). However, the interaction term in this model
was not statistically significant.
Fig. 5. The critical response times and standard errors of egocentric and other-centric responders at 90 and 180 degrees of partner rotation, and sharedperspective response times at 0 degrees of rotation. Between 0–90°, and 90–180°, other-centric responders take more time, whereas egocentric responders
increase from 0° to 90°. However, other-centric responders are generally slower than egocentric responders (Section 4.3.2), including shared-perspective
response times at 0 degrees of rotation (Section 4.3.4). Furthermore, other-centric critical and shared-perspective response times at 90 degrees of rotation
are nearly identical (Section 4.3.4).
Please cite this article in press as: Duran, N. D., et al. Listeners invest in an assumed other’s perspective despite cognitive cost. Cognition
(2011), doi:10.1016/j.cognition.2011.06.009
13
N.D. Duran et al. / Cognition xxx (2011) xxx–xxx
Table 3
Study 3: Means and standard errors for directional shifts, maximum deviation, and acceleration components at 0, 90, and 180 degrees of rotation.
Rotation
0
90
180
Directional shifts
Maximum deviation
Acceleration components
Ego
Other
Ego
Other
Ego
Other
1.14 (.05)
1.16 (.07)
1.29 (.06)
1.41 (.09)
1.54 (.12)
1.73 (.10)
9.08 (1.14)
9.02 (1.13)
9.73 (.93)
16.22 (2.25)
24.01 (3.13)
21.64 (2.25)
9.23 (.22)
9.71 (.32)
10.39 (.27)
11.56 (.47)
13.00 (.69)
12.71 (.59)
4.3.3.2. Maximum deviation. Both main effects and the
interaction were statistically significant. With further analysis of the interaction, F(2, 1339) = 3.34, p = .04, simple
main effects for degree of rotation at responder type show
that only other-centric was statistically significant,
F(2, 1330.17) = 4.88, p = .008, with the greatest deviation
occuring from 0 to 90 degrees of rotation (mean difference = 7.79 pixels), p = .003, d = .30. Simple main effects
for responder type at degree of rotation also show that
other-centric responders had greater deviation toward
the competitor response at 0° (mean difference = 7 pixels),
F(1, 87.31) = 4.97, p = .03, d = .30; at 90° (mean difference = 15 pixels), F(1, 117.61) = 18.80, p < .001, d = .66;
and
at
180°
(mean
difference = 12
pixels),
F(1, 95.14) = 13.22, p < .001, d = .56.
4.3.3.3. Acceleration components. Again, both main effects
were statistically significant, but the interaction term was
not. Interpreting the main effects, there is an overall increase from 0° to 90° (.96 components) and from 90° to
180° (.67 components), F(2, 1168.43) = 16.56, p < .001.
There is also a greater number of acceleration components
for other-centric responders (12.76 components) than egocentric responders (9.78 components), F(1, 51.30) = 7.38,
p = .009.
4.3.4. Perspective-taking during shared trials
As reported in Section 4.3.2 above, and shown in Fig. 5,
at 0 degrees of rotation the response times for othercentric responders were 339 ms slower than egocentric
responders. For shared trials at 90 degrees of rotation,
other-centric responders are again slower, now by
365 ms, F(1, 52.35) = 8.55, p = .005, d = .54. Furthermore,
there were no statistically significant differences between
shared and critical trials at 90 degrees of rotation. Taken
together, these results provide further evidence that
other-centric responders are continuing to mentally rotate
even when it is not necessary to do so.
4.4. Brief discussion
In Studies 1 and 2, participants interacted with a recorded speaker who was unable to share in the work of
forging mutual understanding. In the current study, the
same speaker now has the potential to be a viable contributor who shares the participant’s goals of collaboration. If
participants are sensitive to this information, and are willing to act on it, then previous demands that led to othercentricism are largely diminished. As such, the expected
result is a greater percentage of egocentric responders. Indeed, this was found. There was a significant increase of
egocentricism when compared to Studies 1 and 2. But like
these earlier studies, there were still some who chose to be
other-centric – nearly a third of the participants. Like the
egocentric participants in Study 1 who appeared to have
a stricter criterion for when they would take another’s perspective, the other-centric participants in this study might
also need more evidence that the speaker is acting with
them in mind (thereby allowing the participants to be egocentric). Interestingly, the cognitive ease associated with
taking an egocentric perspective does not seem to matter.
The other-centric responders still continued to expend
more cognitive effort, as shown in the mental rotation
analysis for response time, complexity, and competition.
Moreover, this effort was sustained over trials, even when
visual perspectives between participant and speaker were
shared.
5. Additional analysis: Right/left vs. back/front trials
In this paper, we have focused on spatial descriptions
that occurred in two referential systems, one where objects
are situated in terms of one’s own bodily coordinates, and
the other in which an external agent’s coordinates are
adopted. The descriptions were based on a common set
of referential terms, ‘‘right,’’ ‘‘left,’’ ‘‘front,’’ and ‘‘back,’’ that
could be used to distinguish folders relative to the viewpoint of either agent. For the right/left terms, a folder occured in conjunction with either an agent’s right or left sides,
thus intrinsic side was sufficient for identification. For
back/front trials, an alternative conceptualization is
needed, as folders were always to an agent’s front. In these
trials, participants are likely assigning the folder nearest an
agent (either themselves or their partner) as being the one
in ‘‘front,’’ or by imposing an intrinsic ‘‘front’’ or ‘‘back’’ to
the folder itself. Either way, different conceptual representations might lead to different processing costs, with previous research suggesting that back/front referential terms
are easiest to process (e.g., Franklin & Tversky, 1990).
Our goal in this section is to determine whether the
processing costs of right/left trials substantially vary from
back/front trials; and if so, how this might change our current interpretations of cognitive effort in perspectivetaking. To start, we combined the data from Studies 1 to
3 and plotted the response times of egocentric and othercentric participants’ right/left and back/front trials (which
divides the dataset into two equal parts). Based on visual
inspection alone (see Fig. 6), the bulk of the cognitive work
appears to be carried by the right/left trials. For these referential terms, there is considerable separation between
egocentric and other-centric response times, and when
compared to egocentric participants, the other-centric par-
Please cite this article in press as: Duran, N. D., et al. Listeners invest in an assumed other’s perspective despite cognitive cost. Cognition
(2011), doi:10.1016/j.cognition.2011.06.009
14
N.D. Duran et al. / Cognition xxx (2011) xxx–xxx
Fig. 6. The right/left and back/front critical response times of egocentric and other-centric responders at 90 and 180 degrees of partner rotation, and
shared-perspective response times at 0 degrees of rotation.
Table 4
For Studies 1–3, with back/front and left/right trials, the average increase of
response time across degree of rotation (linear slope) and response time
differences between egocentric and other-centric responders.
Table 5
For Studies 1–3, with just left/right trials, the average increase of response
time across degree of rotation (linear slope) and response time differences
between egocentric and other-centric responders.
All trials
Right/left trials
Study 1
Slope (b) Other
Slope (b) Ego
0 (Other–Ego)
90 (Other–Ego)
180 (Other–Ego)
Study 2
⁄⁄⁄
171 ms
76 ms ⁄⁄⁄
183 ms
232 ms ⁄
361 ms ⁄⁄⁄
Note: statistical significance is coded as:
132 ms
46 ms
203 ms
142 ms
380 ms
⁄
p < .05,
Study 3
⁄⁄⁄
⁄⁄
⁄⁄
p < .01,
Study 1
⁄⁄⁄
195 ms
87 ms ⁄⁄⁄
339 ms ⁄⁄
468 ms ⁄⁄⁄
537 ms ⁄⁄⁄
⁄⁄⁄
p < .001.
ticipants show a much steeper rise in response times
across degree of rotation. Conversely, the back/front trials
for egocentric and other-centric responders are essentially
the same, with minimal increases over degree of rotation.
The pattern in Fig. 6 does not alter our earlier claims
that there are greater cognitive costs in being other-centric, and that these costs are marked by time-consuming
visuospatial transformations. However, an update is
needed. When mental transformations do occur, they do
so between the intrinsic ‘‘right’’ and ‘‘left’’ sides of speaker
and participant. This finding also suggests that by combining right/left and back/front trials, we have inadvertently
weakened the critical effects reported in Studies 1–3. To
examine this possibility, we compare our previous findings
with a new round of analysis using just right/left trials.
Rather than reporting the entirety of the additional statistical testing, we simply report the response time differences and p values. For example, Table 4 shows a
summary of the results from Studies 1 to 3, including the
average increase of response times across degree of rotation (e.g., slope (b) Other), and the difference in response
times between egocentric and other-centric responders
(e.g., 0 (Other–Ego)). Asterisks by each response time indi-
Slope (b) Other
Slope (b) Ego
0 (Other–Ego)
90 (Other–Ego)
180 (Other–Ego)
Study 2
⁄⁄⁄
271 ms
128 ms⁄⁄⁄
228 ms⁄
438 ms⁄⁄
620 ms⁄⁄⁄
Note: statistical significance is coded as:
Study 3
⁄⁄⁄
257 ms
140 ms⁄⁄⁄
210 ms
318 ms⁄
703 ms⁄⁄⁄
⁄
p < .05,
⁄⁄
p < .01,
386 ms⁄⁄⁄
104 ms⁄⁄⁄
377 ms⁄⁄
724 ms⁄⁄⁄
990 ms⁄⁄⁄
⁄⁄⁄
p < .001.
cate the level of significance. Using the same analytical approach as before, Table 5 provides a snapshot of the results
when back/front trials are removed. By comparing these
two tables, it is clear that our earlier interpretations from
Studies 1 to 3 still hold. Moreover, an even stronger case
can be made given the response times in Table 5 (without
back/front trials) greatly exceed those shown in Table 4
(with back/front trials),7 and are still significant despite
the considerable loss of statistical power.
Given the improvement for response times, we now
have reason to believe that the use of ‘‘back’’ and ‘‘front’’
directions may have also obscured critical effects in the
trajectory measures. Because trajectory measures are
much more sensitive to subtle changes in response movements, any noise introduced by the back/front trials is
likely to be amplified. This sensitivity might explain why
the trajectory measures across the three studies were not
as robust as the response time results. By removing back/
front trials, we expect the trajectory measures to provide
7
These simple effects analyses are based on overall models with
statistically significant interaction terms (p < .05).
Please cite this article in press as: Duran, N. D., et al. Listeners invest in an assumed other’s perspective despite cognitive cost. Cognition
(2011), doi:10.1016/j.cognition.2011.06.009
15
N.D. Duran et al. / Cognition xxx (2011) xxx–xxx
Table 6
For select trajectory measures, the average increase for each measure across degree of rotation (linear slope) and differences between egocentric and othercentric responders (right/left trials only). The statistically significant interactions between rotation and responder type are also reported.
Right/left trials
Study 2
Interaction
Slope (b) Other)
Slope (b) Ego
0 (Other–Ego)
90 (Other–Ego)
180 (Other–Ego)
Study 3
Maximum deviation
Acceleration components
Directional shifts
Acceleration components
F(2,993.64) = 2.84, p = .05
5.57⁄⁄⁄
1.62
2.96
19.81⁄⁄
16.72⁄
F(2,988.63) = 5.49, p = .004
1.43⁄⁄⁄
0.09
2.05⁄
2.12⁄
4.37⁄⁄⁄
F(2,733.76) = 5.55, p = .004
0.35⁄⁄⁄
0.07
0.23
0.35
0.83⁄⁄⁄
F(2,731.50) = 11.76, p < .001
2.33⁄⁄⁄
0.27
2.49
4.92⁄⁄⁄
6.49⁄⁄⁄
Note: statistical significance is coded as:
⁄
p < .05,
⁄⁄
p < .01,
⁄⁄⁄
p < .001.
better support for increased complexity and competition in
other-centric response movements.
A new series of analysis was conducted for each of the
three trajectory measures for each of the three studies.
The general trends found in the earlier studies were again
replicated, albeit with increased or similar values for
other-centric responders relative to egocentric responders.
Because of the large number of additional analyses, and given that they are consistent with our previous conclusions,
we highlight only those results that had the largest gains.
In these earlier analyses, there were four (out of nine) trajectory measure models that showed no interaction between degree of rotation and responder type. In the new
analysis, the interaction term for each model is now statistically significant (see Table 6). A follow-up analysis across
degree of rotation reveals evidence of mental rotation for
just other-centric responders. And importantly, follow up
analysis comparing responder type at degree of rotation
(i.e., ground position) shows statistically significant differences between egocentric and other-centric responders at
several positions.
6. General discussion
In the introduction, we reviewed the extensive work on
the role of perspective-taking in spatial memory and other
forms of linguistic and navigational behavior. Yet, in a social context, few have looked at the listener’s natural tendencies to disambiguate perspectives in spatial
description. In the studies presented here, we created a
simple communicative context that required participants
to select one of two referents based on a simulated partner’s requests. The scenario is modeled after a hypothetical
situation where a person, standing next to a table, is asked
by someone else for one of two folders that are laid out on
the table. If the person making the request says, ‘‘Give me
the one on the right’’, and is directly across from the standing person (i.e., addressee), it is unclear if the addressee
should assume their own ‘‘right’’ (an egocentric interpretation) or the requester’s ‘‘right’’ (an other-centric
interpretation).
In Study 1, it is during these open-ended trials, with
loose constraints for how to disambiguate, that we evaluated participants’ spontaneous perspective-taking behavior. We found that more than half of the participants
consistently provided an other-centric response, although
this did not statistically differ from number of egocentric
responders. Even so, the other-centric perspective was
chosen despite a clear processing advantage for egocentric
responding, as evidenced by faster response times and less
complexity in egocentric response movements.
In Study 2, we sought to better grasp the driving force
that might influence participants to be other-centric. We
hypothesized that such responding was largely motivated
by attributions about the simulated partner’s ability (or
inability) to contribute to mutual understanding. However,
there is always the possibility that the artificial nature of
the task also places unintended task demands on a participant. To rule this out, we kept the same design as Study 1,
but added a single cue that was directed at influencing
partner-specific attributions. Here, the critical attributional
cue was a slight change to the instructions, informing participants that their partner does not know where they are
located. This information makes it impossible for the partner to consider any perspective other than his own. Following the goals of collaboration, participants who are
sensitive to this more explicit information are likely to take
on the increased burden of interpreting from the one perspective that is guaranteed to be understood, namely, the
simulated partner’s. Compared to Study 1, the result
should be an increased number of other-centric responders, and a continued cognitive cost of responding othercentrically. As expected, this predicted outcome was
confirmed.
Following the same rationale as in Study 2, we sought to
manipulate a single attributional cue that would now bias
response behavior toward egocentrism. By doing so, a
stronger case could be made about the role of partner-specific attributions in influencing perspective-taking behavior. In Study 3, the same design as Study 1 was used, but
this time we removed the constraint of having to interact
with a simulated partner. Participants now believed that
they were receiving directions from a real person – a person who might share responsibility in working toward mutual understanding. Moreover, because this ‘‘real’’ partner
was the one making the requests, and did not correct any
of the participant’s responses, we expected participants
to direct most of the responsibility toward the partner. If
so, participants are likely to assume the partner is acting
other-centrically, thereby choosing egocentric behavior
for themselves. Indeed, compared to Study 1, this predic-
Please cite this article in press as: Duran, N. D., et al. Listeners invest in an assumed other’s perspective despite cognitive cost. Cognition
(2011), doi:10.1016/j.cognition.2011.06.009
16
N.D. Duran et al. / Cognition xxx (2011) xxx–xxx
tion was supported. We also found a modest number of
other-centric responders who were apparently unswayed
by the new attributional cue.
In all three studies, the other-centric responders were
willing to invest in behavior that was cognitively challenging. In similar research using a spatial perspective-taking
task (Schober, 1993; Tversky & Hard, 2009), it is assumed
that the increased cognitive effort is the result of mentally
rotating to the visual perspective of another. However,
without direct evidence to support this claim, it is unclear
if participants are in fact embodying the participant’s point
of view. The experimental task presented here allows us to
test this supposition. If participants are mentally rotating,
then the greater the offset of one’s own visual perspective
from that of a partner, the longer it will take to rotate to
the partner’s perspective. And indeed, this was found for
three positions that were offset by an equivalent and
increasing magnitude of distance. Other-centric responders were fastest when there was no rotation (shared perspective), but were significantly slower when shared
visual perspective was offset by 90°, and even slower when
shared visual perspective was offset by an additional 90°.
The extent by which these responses were slower also exceeded any slowdowns exhibited by egocentric
responders.
These findings suggest that other-centric participants
are anchoring to an egocentric frame of reference and making adjustments relative to this frame, a process Zacks and
Michelon (2005) calls a visuospatial perspective transformation. However, an egocentric visual anchor is not necessarily isomorphic to a socially-motivated egocentric bias. That
is, we cannot say that participants are first employing and
inhibiting an egocentric bias before considering the perspective of their partner. Our data suggest another possibility. During trials at 0 degrees of rotation, the
participant’s visual perspective is equivalent to their partner’s perspective. These shared visual perspective trials require no mental rotation; thus, if the other-centric
responder has been inhibiting an egocentric bias in making
a rotation, they are now free from having to do so. As a
consequence, response times should be at least equivalent
to the egocentric responder. But this is not the case. Othercentric responders are still significantly slower in shared
trials. If a strong egocentric bias is not at play, what then
causes the slower response times? We argue that the increased time results from an other-centric responder who
is still considering what his partner knows, even when visual perspective is aligned. Contrary to Herrmann (1988),
Herrmann, Burkle, and Nirmaier (1987) early assertions
(as reported by Schober (1993)), participants will indeed
‘‘step into the shoes’’ of their partner even with a shared
vantage point. Once a conceptual perspective has been
adopted, participants are consistent in the way they apply
this perspective in interpreting spatial referents. In a sense,
participants are still metaphorically ‘‘rotating,’’ even when
visual perspectives are aligned.
These findings also have direct implications for the current debate about exactly how partner-specific information is processed during communication. Our claims
coincide with Brennan and colleagues’ ‘‘one-bit’’ theory
that simple yet significant attributional cues can have an
immediate effect on behavioral outcomes (e.g., Brennan &
Hanna, 2009; Galati & Brennan, 2010), particularly when
the cues are easily tracked and introduced early on in the
communicative context (Bortfeld & Brennan, 1997; Hanna
& Tanenhaus, 2004; Tversky & Hard, 2009). Participants
commit wholesale to the elicited perspective (egocentric
for Study 3 and other-centric for Studies 1 and 2). Moreover, the use of attributional cues appears to be consistent
with Clark and Wilkes-Gibbs’ (1986) principle of least collaborative effort, whereby the motivation to consider another’s perspective is rooted in a shared responsibility to
ensure mutual understanding.
Our results can also be contrasted with another account
of perspective-taking behavior that posits language users
will consider another’s perspective only when a primary
egocentric preference becomes unsustainable (Horton &
Keysar, 1996; Pickering & Garrod, 2004; Shintel & Keysar,
2009). In this view, when there is a communication breakdown, or a breakdown is likely to occur, speakers (and presumably listeners) will engage in repair strategies that
require a more costly other-centric orientation. In the current study, we strove to create a scenario where such repair strategies would be unnecessary. Because a partner’s
trial-to-trial directions signaled no obvious confusion, participants could respond egocentrically without penalty or
communicative mishap. Furthermore, in responding egocentrically, the task could be completed faster – a primary
argument for why an egocentric response should be preferred. And given that the social partner only simulates
the intentions and presence of an actual person, such task
demands could be considered biased toward egocentric
responding (see Gerrig, Brennan, & Ohaeri, 2000 for discussion). Yet, despite these contextual influences that arguably favor an egocentric orientation, participants were
still willing to act on the attributional needs of an assumed
social agent, revealing high rates of other-centric responding in production, as shown by Tversky and Hard (2009),
and even here where comprehension is involved.
It also appears that when participants do act egocentrically, particularly when the attributional cues support an
egocentric perspective (Study 3), there is some evidence
that participants are still tracking information about their
partner. As we reported earlier, response times do increase
as a partner’s position is offset from 0° to 90° and 90° to
180°. Although this increase is far from the required time
to make a visuospatial transformation, it does suggest a
low-level awareness of conflicting perspectives. This possibility is weakly supported with the egocentric responders
in Studies 1 and 2. But because response times in these
studies increase from 0 to 90, and are maintained between
90 and 180, it is more likely that responders are merely
noticing the locational shift of the recorded speaker from
their own vantage point. Thus, an attentional distraction
leads to the slight increase of response time, which indicates, at the very least, that egocentric responders are engaged in the task.
Another notable finding is that the high rate of unsolicited other-centric responding is comparable to previous
findings (e.g., Galati & Brennan, 2010; Horton & Gerrig,
2002; Schober, 1993). Yet, in these studies, there are production- and task-based constraints that may make an
Please cite this article in press as: Duran, N. D., et al. Listeners invest in an assumed other’s perspective despite cognitive cost. Cognition
(2011), doi:10.1016/j.cognition.2011.06.009
N.D. Duran et al. / Cognition xxx (2011) xxx–xxx
other-centric response more likely. For example, participants might be told ahead of time to consider the perspective of an addressee, or to design utterances in the service
of conversational pressures (e.g., to establish common
ground), or be allowed to negotiate and receive feedback
from their partner. Of course, in the current research, we
took a different tack by evaluating perspective-taking
behavior as simple comprehension. As with Tversky and
Hard’s (2009) paradigm, we imposed the most minimal
of task requirements for eliciting perspective choice – at
least for Study 1. When comparing these similarly matched
studies, 53% of our participants and 20% of Tversky and
Hard’s participants freely chose to be other-centric. Importantly, this difference in response rate also brings up a critical distinction between the two studies. Here,
perspective-taking behavior was allowed to stabilize over
a series of trials. As in previous research involving a real
or simulated social partner, perspective choice is sometimes gradually worked out in the early moments of interaction (Bard & Aylett, 2001; Horton & Keysar, 1996; Roche,
Dale, & Kreuz, 2010). Indeed, in our Study 1, 22 other-centric participants, or 51% overall, had between 1 and 3
short-lived ‘‘egocentric’’ responses in the initial trials. If
we had assessed perspective-taking behavior after a single
trial, these respondents would have been misclassified as
egocentric.
There are other differences that make our particular
task promising for further exploration. One of these is
tracking the temporal dynamics of social perspective-taking. In the present studies, we evaluated simple computer
mouse movements which revealed that other-centric
responding takes more time to execute, and that the executed movement is characterized by greater directional
shifts and decelerations. These latter findings suggest the
presence of a processing interference that arises when
adopting another’s frame of reference. Presumably, a major
source of interference is from one’s own activated frame of
reference (Brown-Schmidt, 2009). Although multipleframe activation and the primacy of a single activation
has received a great deal of attention in ambiguous scene
interpretation (Carlson-Radvansky & Irwin, 1994;
Carlson-Radvansky & Logan, 1997), it is less clear with
social perspective-taking. One issue that we examined in
greater detail is whether reference frames are activated
sequentially or in parallel (Taylor et al., 1999). If in parallel,
we expected to find competition effects during the
moment-by-moment changes during the response movement itself. These competition effects can be expressed in
paradigms like ours where a competitor and response target are visually co-present (Song & Nakayama, 2009; Spivey & Dale, 2006). Here, we provided a simple measure
of how much the movement is displaced toward a competitor en route to a target. For other-centric responders, we
found that movements towards a folder designating an
other-centric frame were shifted towards a folder designating an egocentric frame. This result suggests both
frames were simultaneously activated, but with the
other-centric response eventually winning out.
Greater evidence for competition effects, as well as processing costs, can also be had with situational factors that
are easily implemented in this task. For example, we can
17
evaluate perspective-taking behavior when there are
greater cognitive demands, as in introducing time constraints to respond (as in Bard et al., 2007). We might also
find that with stronger attributional cues that favor an
other-centric perspective, egocentric responders will have
significant difficulty in inhibiting an other-centric frame
of reference, perhaps even from the earliest stages of processing. And importantly, this basic task allows us to systematically identify pragmatically relevant cues that
might sway the activation and selection of egocentric or
other-centric perspectives. It also opens up the possibility
of exploring any individual differences that might interact
with these factors.
On a final theoretical note, a further possible interpretation of our results is that so-called ‘‘default’’ processes are
only one part of the explanatory story of adaptive abilities
during interaction. As Schober (1993) demonstrated, the
coordinative process of interaction permits rich patterns
of adaptation that emerge when one looks for them in
appropriately interactive contexts (see also Brennan &
Hanna, 2009; Brown-Schmidt, Gunlogson, & Tanenhaus,
2008). Yet, we have shown that even simulated social contexts can induce complete investment in more ‘‘difficult’’
response strategies. This is not simply a matter of monitoring, because the 0-degree trials in our studies still show
that participants have embraced the strategy of trialby-trial transformation for the benefit of an absent but
assumed partner. So ‘‘defaults’’ are transformed in
socially-embedded contexts, and the ‘‘tables are turned,’’
so to speak, because many participants have now organized their interpretive tendencies outside themselves,
into the other-centric frame. Our data suggest that a new
default, adapted for the benefit of the task, can become
that external frame. This suggests that theories of perspective-taking, and partner-specific adaptation in general,
need to focus attention on stable strategies in more taskspecific ways, rather than seeking generalizations that
may overreach, and miss the natural human capacity to
adapt cognitive processes in novel circumstances.
Acknowledgments
The research was supported by NSF Grant BCS-0720322
to the second author, and NSF Grant BCS-0826825 to the
second and third authors.
References
Amorim, M. A. (2003). ‘‘What is my avatar seeing?’’ The coordination of
‘‘out-of-body’’ and ‘‘embodied’’ perspectives for scene recognition
across views. Visual Cognition, 10, 157–199.
Baayen, R. H. (2008). Analyzing linguistic data: A practical introduction to
statistics in R. New York: Cambridge University Press.
Baayen, R. H., Davidson, D. J., & Bates, D. M. (2008). Mixed-effects
modeling with crossed random effects for subjects and items. Journal
of Memory and Language, 59, 390–412.
Bard, E. G., Anderson, A. H., Chen, Y., Nicholson, H. B. M., Havard, C., &
Dalzel-Job, S. (2007). Let’s you do that: Sharing the cognitive burdens
of dialogue. Journal of Memory and Language, 57, 616–641.
Bard, E. G., & Aylett, M. P. (2001). Referential form, word duration, and
modeling the listener in spoken dialogue. In J. Moore & K. Stenning
(Eds.), Proceedings of the 23rd Annual Meeting of the Cognitive Science
Society. Austin, TX: Cognitive Science Society.
Please cite this article in press as: Duran, N. D., et al. Listeners invest in an assumed other’s perspective despite cognitive cost. Cognition
(2011), doi:10.1016/j.cognition.2011.06.009
18
N.D. Duran et al. / Cognition xxx (2011) xxx–xxx
Black, J. B., Turner, T. J., & Bower, G. H. (1979). Point of view in narrative
comprehension, memory, and production. Journal of Verbal Learning
and Verbal Behavior, 18, 187–198.
Bortfeld, H., & Brennan, S. E. (1997). Use and acquisition of idiomatic
expressions in referring by native and non-native speakers. Discourse
Processes, 23, 119–147.
Branigan, H. P., Pickering, M. J., Pearson, J., & McLean, J. F. (2010).
Linguistic alignment between people and computers. Journal of
Pragmatics, 42, 2355–2368.
Brennan, S.E. (1990). Speaking and providing evidence for mutual
understanding.
Unpublished
doctoral
dissertation,
Stanford
University, Stanford, CA.
Brennan, S. E., & Hanna, J. E. (2009). Partner-specific adaptation in dialog.
Topics in Cognitive Science, 1, 274–291.
Brown-Schmidt, S. (2009). Partner-specific interpretation of maintained
referential precedents during interactive dialog. Journal of Memory
and Language, 61, 171–190.
Brown-Schmidt, S., Gunlogson, C., & Tanenhaus, M. K. (2008). Addressees
distinguish shared from private information when interpreting
questions during interactive conversation. Cognition, 107, 1122–1134.
Brunyé, T. T., Rapp, D. N., & Taylor, H. A. (2008). Representational
flexibility and specificity following spatial descriptions of real-world
environments. Cognition, 108, 418–443.
Carlson, L. A. (1999). Selecting a reference frame. Spatial Cognition and
Computation, 1, 365–379.
Carlson, L. A., & Covey, E. S. (2005). How far is near? Inferring distance from
spatial descriptions. Language and Cognitive Processes, 20, 617–631.
Carlson-Radvansky, L. A., & Irwin, D. E. (1994). Reference frame activation
during spatial term assignment. Journal of Memory and Language, 33,
646–671.
Carlson-Radvansky, L. A., & Logan, G. D. (1997). The influence of reference
frame selection on spatial template construction. Journal of Memory
and Language, 37, 411–437.
Clark, H. H. (1996). Using language. Cambridge: Cambridge University
Press.
Clark, H. H., & Krych, M. A. (2004). Speaking while monitoring addresses
for understanding. Journal of Memory and Language, 50, 62–81.
Clark, H. H., & Schaefer, E. F. (1989). Contributing to discourse. Cognitive
Science, 13, 259–294.
Clark, H. H., & Wilkes-Gibbs, D. (1986). Referring as a collaborative
process. Cognition, 22, 1–39.
Dale, R., & Duran, N. D. (in press). The cognitive dynamics of negated
sentence verification. Cognitive Science.
Dale, R., Duran, N. D., & Morehead, R. (submitted for publication).
Prediction in statistical learning, and implications for the explicit/
implicit divide.
Epley, N., Keysar, B., Van Boven, L., & Gilovich, T. (2004). Perspective
taking as egocentric anchoring and adjustment. Journal of Personality
and Social Psychology, 87, 327–339.
Fitneva, S. A., & Song, Y. (2009). The comprehension of ‘‘left’’ and ‘‘right’’ in
a referential communication task. In N. A. Taatgen & H. van Rijn (Eds.),
Proceedings of the 31st Annual Conference of the Cognitive Science
Society (pp. 2687–2691). Austin, TX: Cognitive Science Society.
Frank, M. C., & Gibson, E. (2011). Overcoming memory limitations in rule
learning. Language, Learning, & Development, 7, 130–148.
Franklin, N., & Tversky, B. (1990). Searching imagined environments.
Journal of Experimental Psychology: General, 119, 63–76.
Fussell, S. R., & Krauss, R. M. (1992). Coordination of knowledge in
communication: Effects of speakers’ assumptions about what others
know. Journal of Personality and Social Psychology, 62, 378–391.
Galati, A., & Brennan, S. E. (2010). Attenuating information in spoken
communication: For the speaker, or for the addressee? Journal of
Memory and Language, 62, 35–51.
Gerrig, R. H., Brennan, S. E., & Ohaeri, J. O. (2000). What can we conclude
from speakers behaving badly? Discourse Processes, 29, 173–178.
Hanna, J. E., & Tanenhaus, M. K. (2004). Pragmatic effects on reference
resolution in a collaborative task: Evidence from eye movements.
Cognitive Science, 28, 105–115.
Herrmann, T. (1988). Partnerbezogene Objektlokalisation – ein neues
sprachpsychologisches Forschungsthema. Bericht Nr. 25, Arbeiten der
Forschergruppe ‘‘Sprechen und Sprachverstehen im sozialen
Kontext’’, Heidelberg/Mannheim.
Herrmann, T., Burkle, B., & Nirmaier, H. (1987). Zur horerbezogenen
Raumreferenz: Horerposition und Lokalisationsaufwand. Bericht Nr. 12,
Arbeiten der Forschergruppe ‘‘Sprechen und Sprachverstehen im
sozialen Kontext’’, Heidelberg/Mannheim.
Horton, W. S., & Gerrig, R. J. (2002). Speaker’s experiences and audience
design: Knowing when and knowing how to adjust utterances to
addressees. Journal of Memory and Language, 47, 589–606.
Horton, W. S., & Gerrig, R. J. (2005). The impact of memory demands on
audience design during language production. Cognition, 96, 127–142.
Horton, W., & Keysar, B. (1996). When do speakers take into account
common ground? Cognition, 59, 91–117.
Isaacs, E. A., & Clark, H. H. (1987). References in conversation between
experts and novices. Journal of Experimental Psychology: General, 116,
26–37.
Kittur, A., Chi, H., & Suh, B. (2008). Crowdsourcing user studies with
Mechanical Turk. ACM Press. CHI 2008.
Lee, P. U., & Tversky, B. (2005). Interplay between visual and spatial: The
effect of landmark descriptions on comprehension of route/survey
spatial descriptions. Spatial Cognition & Computation, 5, 163–185.
Levelt, W. J. (1989). Speaking: From intention to articulation. Cambridge,
MA: MIT Press.
Mainwaring, S. D., Tversky, B., Ohgishi, M., & Schiano, D. J. (2003).
Descriptions of simple spatial scenes in English and Japanese. Spatial
Cognition & Computation: An Interdisciplinary Journal, 3, 3–42.
doi:10.1207/S15427633SCC0301_2.
McNamara, T. P., Sluzenski, J., & Rump, B. (2008). Human spatial memory
and navigation. In J. Byrne
(Ed.). Learning and memory: A
Comprehensive reference (Vol. 2, pp. 157–178). Oxford: Elsevier.
Michelon, P., & Zacks, J. M. (2006). Two kinds of visual perspective taking.
Perception & Psychophysics, 68, 327–337.
Miller, G. A., & Johnson-Laird, P. N. (1976). Language and perception.
Cambridge: Cambridge University Press.
Mou, W., Liu, X., & McNamara, T. P. (2009). Layout geometry in encoding
and retrieval of spatial memory. Journal of Experimental Psychology:
Human Perception and Performance, 35, 83–93. doi:10.1037/00961523.35.1.83.
Munro, R., Bethard, S., Kuperman, V., Tzuyin Lai, V., Melnick, R., & Potts, C.,
et al. (2010). Crowdsourcing and language studies: the new
generation of linguistic data. In Proceedings of the NAACL HLT 2010
Workshop on Creating Speech and Language Data with Amazon’s
Mechanical Turk, Los Angeles, CA.
Newman-Norlund, S. E., Noordzij, M. L., Newman-Norlund, R. D., Volman,
I. A. C., de Ruiter, J. P., Hagoort, P., et al. (2009). Recipient design in
tacit communication. Cognition, 111, 46–54.
Perkins, L., & Milroy, L. (1997). Sharing the communicative burden: A
conversation-analytic account of aphasic/non-aphasic interaction.
Multilingua, 16, 199–215.
Pickering, M. J., & Garrod, S. (2004). Toward a mechanistic psychology of
dialogue. Behavioral and Brain Sciences, 27, 169–226.
Presson, C. C., & Montello, D. R. (1994). Updating after rotational and
translational body movements: Coordinate structure of perspective
space. Perception, 23, 1447–1455.
Quene, H., & van den Bergh, H. (2008). Examples of mixed-effects
modeling with crossed random effects and with binomial data.
Journal of Memory and Language, 59, 413–425.
Roche, J. M., Dale, R., & Kreuz, R. J. (2010). The avoidance of ambiguity
during conversation: More than mere priming or mimicry? In S.
Ohlsson & R. Catrambone (Eds.), Proceedings of the 32nd Annual
Meeting of the Cognitive Science Society. Austin, TX: Cognitive Science
Society.
Samson, D., Apperly, I. A., Braithwaite, J. J., Andrews, B. J., & Bodley Scott, S.
E. (2010). Seeing it their way: Evidence for rapid and involuntary
computation of what other people see. Journal of Experimental
Psychology: Human Perception and Performance, 36, 1255–1266.
Schober, M. F. (1993). Spatial perspective-taking in conversation.
Cognition, 47, 1–24.
Schober, M. F. (1995). Speakers, addressees, and frames of reference.
Whose effort is minimized in conversations about locations?
Discourse Processes, 20, 219–247.
Schober, M. F. (1998). Different kinds of conversational perspectivetaking. In S. R. Fussell & R. J. Kreuz (Eds.), Social and cognitive
psychological
approaches
to
interpersonal
communication
(pp. 145–174). Mahwah, NJ: Lawrence Erlbaum.
Schober, M. F. (2009). Spatial dialogue between partners with
mismatched abilities. In K. R. Coventry, T. Tenbrink, & J. A. Bateman
(Eds.). Spatial language and dialogue (Vol. 1, pp. 23–39). Oxford:
Oxford University Press.
Shepard, R., & Metzler, J. (1971). Mental rotation of three dimensional
objects. Science, 171, 701–703.
Shintel, H., & Keysar, B. (2009). Less is more: A minimalist account of joint
action in communication. Topics in Cognitive Science, 1, 260–273.
Snow, R., O’Connor, B., Jurafsky, D., & Ng, A. Y. (2008). Cheap and fast – But
is it good? Evaluating non-expert annotations for natural language tasks.
Presented at the proceedings of the conference on empirical methods in
natural language processing. Stroudsburg, PA: Association for
Computational Linguistics.
Please cite this article in press as: Duran, N. D., et al. Listeners invest in an assumed other’s perspective despite cognitive cost. Cognition
(2011), doi:10.1016/j.cognition.2011.06.009
N.D. Duran et al. / Cognition xxx (2011) xxx–xxx
Song, J. H., & Nakayama, K. (2009). Hidden cognitive states revealed in
choice reaching tasks. Trends in Cognitive Sciences, 13(8), 360–366.
Sorokin, A., & Forsyth, D. (2008). Utility data annotation with Amazon
Mechanical Turk. In Presented at the IEEE Computer Society Conference
on Computer Vision and Pattern Recognition Workshops.
Spivey, M. J., & Dale, R. (2006). Continuous dynamics in real-time
cognition. Current Directions in Psychological Science, 15(5), 207–211.
Taylor, H. A., Naylor, S. J., Faust, R. R., & Holcomb, P. J. (1999). ‘‘Could you
hand me those keys on the right?’’ Disentangling spatial reference
frames using different methodologies. Spatial Cognition and
Computation, 1, 381–397. doi:10.1023/A:1010035613419.
Taylor, H. A., & Tversky, B. (1992). Spatial mental models derived from
survey and route descriptions. Journal of Memory and Language, 31,
261–292. doi:10.1016/0749-596X(92)90014-O.
19
Taylor, H. A., & Tversky, B. (1996). Perspective in spatial descriptions.
Journal of Memory and Language, 35, 371–391.
Tversky, B., & Hard, B. M. (2009). Embodied and disembodied cognition:
Spatial perspective-taking. Cognition, 110, 124–129.
Tversky, B., Lee, P., & Mainwaring, S. D. (1999). Why do speakers mix
perspectives? Spatial Cognition and Computation, 1, 399–412.
Wilkes-Gibbs, D., & Clark, H. H. (1992). Co-ordinating beliefs in
conversation. Journal of Memory and Language, 31, 183–194.
Zacks, J. M., & Michelon, P. (2005). Transformations of visuospatial
images. Behavioral and Cognitive Neuroscience Reviews, 4, 96–118.
Zacks, J. M., Vettel, J. M., & Michelon, P. (2003). Imagined viewer and
object rotations dissociated with event-related fMRI. Journal of
Cognitive Neuroscience, 15, 1002–1018.
Please cite this article in press as: Duran, N. D., et al. Listeners invest in an assumed other’s perspective despite cognitive cost. Cognition
(2011), doi:10.1016/j.cognition.2011.06.009